Abstract: Introduction: The World Health Organization states that Post Partum Hemorrhage (PPH) is the leadingcause of mortality and severe maternal morbidity worldwide. Therefore, it becomes essential to study risk factors in order to identify women at risk to improve their clinical management. The aim of this studywas to identify the risk factors associated with primary post-partum hemorrhage and to determine the accuracy of artificial neural networks as a statistical method to identify mayor risk factors. The hypothesis is that artificial neural networks can be a valid forecasting tool for hemorrhage risk at the time of delivery.Methods: We conducted a retrospective study on 305 women with PPH and 305 controls matched by age, analyzing the role of variables that have been identified as possible risk factors. Other ante and intra-partum variables were also taken into consideration in order to identify other risk factors previouslyundetected. Traditional statistical methods and artificial neural networks (ANN) were used to analyze the data. Auto Contractive Maps were used as a neural network system. These have been shown to be capable of identifying hidden, non-linear associations and the strength that these associations have with the different parameters taken into account.Results: The results obtained by the two statistical methods were similar and ANN identified PPH with an overall accuracy of 70%. Variables that were significantly related to PPH overlapped and were mainly related to labor and delivery. Furthermore, neural networks allowed us to identify the impact of each variable. 
Conclusion: Artificial neural networks have proved to be capable of reaching a good sensitivity, specificity, and global accuracy in the identification of more relevant hemorrhage risk factors. The innovative potential of this method is represented by the possibility of identifying a priori women with an increased hemorrhagic risk. This will allow tailoring the management of these women minimizing the consequences that a delivery complicated by hemorrhage involves. These results represent the starting point for the expansion of the retrospective case series and the creation of an antepartum risk score, in addition to a subsequent prospective validation.

Notes:

1University of Milano, Milano, Italy

2Villa Santa Maria Foundation, Tavernerio, Italy